[Bug 26249] add GObject-Introspection bindings

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=26249

--- Comment #20 from Danielle Madeley <danielle.madeley at collabora.co.uk> 2010-04-22 20:14:16 PDT ---
(In reply to comment #19)
> Merge blockers
> ==============
> 
> In "Initial work on generating gobject-introspection":
> 
> > +       LD_LIBRARY_PATH=.libs$${LD_LIBRARY_PATH:+:$$LD_LIBRARY_PATH} \
> 
> If you invoke it using $(top_builddir)/libtool --mode=execute
> $(INTROSPECTION_COMPILER) ... (like we do for valgrind in the regression
> tests), you shouldn't need to set LD_LIBRARY_PATH by hand.

Fixed.

> >   * tp_asv_set_boolean:
> 
> I think it's dangerous to bind these setters into other languages - they make
> assumptions about the memory allocation model of the GHashTable that are not,
> in general, true. (Also, we don't bind tp_asv_new, which is the only documented
> way to create a suitable hash table :-)

Ok. I think we should skip all tp_asv_* functions.

> > -               --pkg dbus-glib-1 \
> 
> Do we want to keep the dbus-glib-1 pkg-config package, even though we've
> dropped the GObject-Introspection-level dependency? I think that would avoid
> needing @DBUS_CFLAGS@?

You're right. Fixed.

> > JS example that retrieves the statuses of valid accounts
> > Connection Manager example
> > Contact list example that makes a low-level D-Bus call
> 
> I don't think we should ship these at all, for now.

Agreed.

> In future, examples that are exemplary should go in examples/client/ (or
> perhaps in examples/client/js/), and non-automated tests should go in
> tests/manual/ or something, to make it very obvious that unlike the rest of the
> tests, they are not run automatically.

Agreed.

> Not merge blockers
> ==================
> 
> In "Initial work on generating gobject-introspection":
> 
> > +       proxy.c _gen/proxy-introspectable.h \
> > +       account.c account.h \
> > +       account-manager.c account-manager.h \
> > +       connection.c connection.h \
> > +       connection-handles.c \
> > +       connection-manager.c connection-manager.h \
> > +       channel.c channel.h \
> > +       handle.c handle.h \
> > +       dbus-daemon.c dbus-daemon.h \
> > +       interfaces.c interfaces.h \
> > +       intset.c intset.h \
> > +       dbus.c dbus.h \
> > +       capabilities.c capabilities.h \
> > +       contact.c contact.h \
> > +       defs.h \
> > +       debug.c debug.h \
> > +       _gen/telepathy-enums.h \
> > +       _gen/telepathy-interfaces.h
> 
> I'd still like a brief status report for each of those files: unchecked /
> completely OK / only foo and bar checked / whatever.

Will build a list.

> It might help to look at the generated .gir and see if it looks sane.
> 
> >     * TpChannel::group-members-changed:
> 
> Why is this already re-annotated in the initial commit? Also, weren't you going
> to (skip) it?

It was annotated to make the original commit compile. Will (skip) instead.

> >    * TpChannel::group-members-changed-detailed:
> 
> Likewise, but without the (skip)ping.

Same reason.

> > @@ -738,8 +738,8 @@ connection_got_contact_attributes (TpConnection *self,
> >   * @user_data: arbitrary user-supplied data
> 
> Why is this already re-annotated in the initial commit? Also, user_data could
> probably be annotated as (closure).

Same reason. This method doesn't actually work anyway.

> > Annotate TpProxy
> 
> Have you checked all of TpProxy's API? If not, "Annotate
> tp_proxy_prepare_async" would seem a more appropriate commit message.

So far everything I've tried is introspected correctly.

> > Annotate TpChannel
> >  342  * tp_channel_borrow_connection:
> 
> Shouldn't methods that duplicate a property be (skip)ped? I consider this
> method to be a "C binding" :-)

I don't know what the G-I policy is here. We should ask.

> > + * Returns: (transfer none): a borrowed reference to the #TpContact:connection
> 
> I think this is a "C binding" and so should be (skip)ped? In fact, most of the
> TpContact stuff falls into that category.
> 
> > + * @contacts: (array length=n_contacts): An array of @n_contacts TpContact
> > + *  objects (this callback is
> >   *  not given a reference to any of these objects, and must call
> >   *  g_object_ref() on any that it will keep), or %NULL on unrecoverable errors
> 
> Shouldn't this have an explicit (transfer none)? Likewise for requested_ids,
> failed_id_errors?

This method doesn't actually work. Hence the bug to replace them with GIO-style
API. I'm not sure whether or not to skip currently broken API or not. If it
doesn't work it's not really part of our API guarantee at the moment anyway.

> > + * tp_get_bus_proxy: (skip)
> 
> All other deprecated functions should probably be (skip)ped, too.

Ok. There were only 3 functions in debug.c not skipped.

> I'd quite like to see a straight Javascript port of the existing client
> examples (inspect-connection and friends) in examples/.

-- 
Configure bugmail: https://bugs.freedesktop.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the QA contact for the bug.